Acquiring Customers via Word-of-Mouth Referrals: A ...

“Acquiring Customers via Word-of-Mouth Referrals: A Virtuous Strategy?” © 2015 Constant Pieters and Aurélie Lemmens; Report Summary © 2015 Marketing Science Institute

MSI working papers are distributed for the benefit of MSI corporate and academic members

and the general public. Reports are not to be reproduced or published in any form or by any means, electronic or mechanical, without written permission.

Marketing Science Institute Working Paper Series 2015

Report No. 15-123

Acquiring Customers via Word-of-Mouth Referrals: A Virtuous Strategy?

Constant Pieters and Aurélie Lemmens

Report Summary

Evidence suggests that customers acquired through seeded word-of-mouth (WOM) campaigns or referral programs have higher margin and lower churn probability than customers acquired in other ways. However, the success of such strategies also depends on the extent to which these customers spread the word among their peers, producing long recommendation “cascades.” In this report, Constant Pieters and Aurélie Lemmens examine this important aspect of customer referral value: To what extent do referred customers pass on the referral they received to others, and what drives this behavior? Based on an analysis of large-scale survey data among U.S. movie viewers, they find that, on average, customers acquired via referrals do not have significantly different referral value than other customers. First, exposure to WOM referrals is non-random; customers who were exposed to WOM referrals are systematically different than customers who were not. Ignoring this self-selection mechanism leads to an overestimation of the effect of referral exposure on referral value. Second, a moderated mediation analysis shows that the mediation of satisfaction explains almost 80% of the total effect of WOM referral exposure on referral value. Referred customers who receive referrals that do not fit their tastes well (badly matched referrals) end up less satisfied than non-referred customers, leading them to refer less in turn. The results suggest that managers should use WOM acquisition strategies cautiously as they may be not as successful in attracting customers with a high referral value as they are in recruiting profitable customers. Moreover, managers should not expect long chains or cascades of referrals as a result of WOM acquisition strategies. Finally, companies should make sure their prospective customers have realistic expectations prior to consumption (for example, by means of information tools), and should encourage referrers to take the recipient’s tastes into account when referring (for example, by means of matching tools). Constant Pieters is a Ph. D. Candidate in Marketing and Aurélie Lemmens is Associate

Professor of Marketing, both at the Tilburg School of Economics and Management, Tilburg

University.

Acknowledgments

The authors received financial support from a N.W.O. VIDI grant. The manuscript benefited from invaluable suggestions by Rik Pieters, Bart Bronnenberg, Marnik Dekimpe, Els Gijsbrechts, and the other members of the marketing department at Tilburg University. The authors acknowledge helpful comments from Ajay Kohli, Michael Haenlein, Marion Debruyne, and participants of the 28th EMAC Doctoral Colloquium in Leuven.


Introduction

Companies show a growing interest in managing the many word-of-mouth (WOM)

interactions between their customers, for instance the product recommendations—further

denoted WOM referrals (Brown and Reingen 1987; Trusov et al. 2009)—customers make to

others. Many firms encourage such WOM referrals with or without monetary incentives, for

example by means of WOM seeding strategies (Godes and Mayzlin 2009; Haenlein and Libai

2013; Hinz et al. 2011). Examples of such seeded WOM are the ones offered by the marketing

company BzzAgent. The company generates WOM communication for its company clients by

encouraging a panel of consumer agents (BzzAgents) to share their opinion about the client’s

products. Customer referral programs or member-get-member campaigns are other popular

examples of WOM acquisition strategies (Garnefeld et al. 2013; Schmitt et al. 2011; Verlegh et

al. 2013). Firms, such as Dropbox or Spotify, reward existing customers for bringing in new

customers, with additional free storage space or a free premium account. WOM is not only seen

as relatively cheap compared to other acquisition tools (e.g., advertising campaigns), it is also

perceived as a more persuasive, credible, and a better targeted source of information (Bone 1995;

Duhan et al. 1997; Murray 1991).

Many academic studies have explored the impact of WOM in fostering new product

adoption and diffusion (Aral and Walker 2014; Nair et al. 2010), or sales (Chevalier and Mayzlin

2006; Godes and Mayzlin 2009). Another stream of research has shown the ability of WOM in

recruiting customers with higher margins and lower churn than customers acquired through other

channels (Schmitt et al. 2011; Villanueva et al. 2008). Firms have experienced these benefits in

practice as well. For instance, the Dropbox refer-a-friend feature increased signups from 100k to

four million users in 15 months.1

Nevertheless, an important question remains whether the customers recruited by referral also

turn out to be good advocates of the firm. In other words, do they pass on the referral they

received to others? This question was raised by Villanueva et al. (2008) who explain that

customers should not only contribute to customer equity of the firm by the stream of their future

cash flows, but also by generating WOM referrals. Likewise, Kumar et al. (2010) recently

pointed out the importance of recruiting customers not only based on their customer lifetime

1 http://www.referralcandy.com/blog/referrals-built-dropbox-empire/


value (CLV), but also taking into account their customer referral value (CRV), that characterizes

the number of WOM referrals a customer makes. They find that the customers with the highest

CRV are typically not the ones with the highest CLV. The success of WOM referral strategies

does not only depend on the CLV of the acquired customers to the firm but also on the extent to

which referred customers continue spreading the good word among their peers (Haenlein and

Libai 2013), which would allow the production of long recommendation cascades or referral

chains (Goel et al. 2012; Leskovec et al. 2007).

Several studies have recently investigated the effect of WOM referral exposure on WOM

referral giving. These are summarized in Table 1 (Tables follow References throughout).2 One

stream of research has found, using individual-level survey data, a positive correlation between

receiving and giving referrals (File et al. 1994; Sheth 1971; Uncles et al. 2013; Von

Wangenheim and Bayón 2006). Recently, Yang et al. (2012) also found a synergy effect between

receiving and giving referrals in that customers get utility from engaging in both actions. These

results have been corroborated by a second stream of studies that has analyzed aggregate time-

series of customer acquisition via referrals (Trusov et al. 2009; Villanueva et al. 2008). They

concluded that customers exposed to WOM referrals contribute to higher referral acquisition

rates in future time periods, compared to customers acquired through traditional marketing

channels. A third stream of work has collected individual-level real-life referral network data

(Goel et al. 2012; Leskovec et al. 2007). In contrast to what other studies would have predicted,

they find that cascades of referrals are rare in practice, suggesting that the effect of referrals on

subsequent referral value might actually be more limited than previously assumed.

This paper readdresses the question of the effect of receiving a referral on a customer's

referral value. Figure 1 (Figures follow References throughout) illustrates our research question

graphically. Circles represent acquired customers and arrows represent customer-to-customer

WOM referrals. Our paper investigates whether the fact that an arrow is pointing at a given

customer (e.g., customer 4, the grey circle) has an influence on the number of arrows originating

from this customer and pointing at others. In case of a large effect, long recommendation

cascades or referral chains should be observed.

2 For conciseness, we exclude from the table the studies that have solely focused on outcomes different than WOM referral giving (e.g., adoption, diffusion, sales or CLV).


In contrast to previous work, this study takes into account the potential self-selection

mechanism by which receiving a product recommendation is a non-random process. Previous

work points out to the possibility that the customers exposed to referrals are intrinsically

different from the customers non-exposed to referrals. For instance, Uncles et al. (2013) note that

customers with a wide circle of friends and a greater interest in products naturally receive and

make more referrals than others. We expect two sources of endogenous selection. First, some

customers may be more likely to activate others to discuss their consumption experiences, while

simultaneously being more inclined to seek social contacts. Similarly, some products are simply

more likely to be talked about (Berger and Schwartz 2011), leading to a potentially spurious

correlation between receiving and giving referrals. From a managerial viewpoint, it is important

to correct for endogeneity because one could erroneously attribute the differences in referral

value to the exposure to referrals, and thus overestimate the effectiveness of WOM acquisition

strategies. To solve this problem, we use an instrumental variable approach in which we specify

a selection equation and allow the error terms of the various equations to be correlated (Skrondal

and Rabe-Hesketh 2004).

In addition to measuring the effect of referral exposure to a customer’s subsequent referral

value, the paper also contributes to the literature by shedding light on the mechanisms driving

this effect. Understanding the underlying mechanisms provides guidelines on how to improve

WOM strategies. To do so, we perform a moderated mediation analysis. First, the model tests the

moderating role of referral match, the extent to which referrals are well-matched to the

recipient’s preferences. Second, we investigate the mediating role of customer satisfaction.

We collect individual referral data via a large-scale survey among a sample of 851

customers in the motion-picture industry. The experiential and intangible nature of movie

consumption makes WOM referral information important in this industry (Murray 1991;

Neelamegham and Jain 1999). Our data indicate the presence of a strong endogenous selection

mechanism through which the customers who received a movie recommendation turn out to be

intrinsically different from the customers who did not. After controlling for endogeneity, we find

that, on average, exposure to WOM referrals has no effect on a customer's referral value.

Interestingly, the results—when ignoring endogeneity—erroneously point to a positive relation

between receiving and giving referrals. The moderated mediation analysis shows that, when

referrals are ill-matched with the recipient’s preferences, the referred customers end up less


satisfied with the referred movie than non-referred customers, leading to a lower referral value.

We find that this mechanism accounts for the majority (almost 80%) of the total effect. We

conclude the paper by providing a number of suggestions for firms to improve the satisfaction of

referred customers and, by this, the success of their WOM referral programs.

The remainder of this paper is structured as follows. In the next section, we provide a

theoretical background to shed light on the relationship between receiving and giving WOM

referrals. We then present the data, the methodology and the empirical application. This paper

closes with a discussion of managerial implications and directions for future research.

Theoretical Framework and Hypotheses

An overview of our theoretical framework is given in Figure 2. We suggest that exposure to

WOM referrals can influence the referral value of customers through the mediating effect of

satisfaction. The positive effect of satisfaction on WOM referral intention and behavior is

relatively well-established (Anderson 1998; De Matos and Rossi 2008). Hence, we verify this

relationship in our empirical analysis. However, a prerequisite for establishing the mediating role

of satisfaction is to show that WOM referral exposure affects satisfaction. The effects of WOM

exposure on satisfaction are less known. Below, we provide a theoretical background to these

effects.

Impact of WOM referral exposure on satisfaction

We define satisfaction as the pleasurable fulfillment of service (Shankar et al. 2003), and use

the expectancy disconfirmation model to develop our arguments (Oliver 1980). The expectancy

disconfirmation model is a comparative model of customer satisfaction that predicts satisfaction

to be the difference between expectations and perceived performance. A customer is predicted to

be satisfied if perceived performance exceeds her expectations, whereas a customer is

dissatisfied when perceived performance falls short of expectations (Oliver 1980). A referral a

customer receives acts as information prior to purchase upon which the customer can base her

expectations regarding performance quality (Anderson and Salisbury 2003; Murray 1991;

Zeithaml et al. 1993). In the motion-picture industry, the role of WOM in forming such

expectations is known to be particularly important because of the experiential and intangible

nature of movie consumption (Murray 1991; Neelamegham and Jain 1999), which makes it hard

for customers to form expectations by other means. Below, we provide several arguments for a


positive effect of WOM referral exposure on expectations, and thus indirectly a negative effect

on satisfaction, consistent with the expectancy disconfirmation framework (Oliver, 1980;

Westbrook, 1987; Zeithaml et al., 1993).

First, the product or service referrals received by customers are positive in nature and made

by customers who positively evaluated the product or service. The more someone enjoyed a

movie, the higher his likelihood to share his positive experience and refer it to friends (Anderson

1998). Through this mechanism, referred customers are likely to receive above-than-average

positive information about a movie, compared to non-referred customers, and so to form high

expectations about the referred movie.

Second, referrals are known to be a particularly valuable source of information for

customers. According to the accessibility/diagnosticity theory (Feldman and Lynch 1988),

information in memory is likely to influence the consumer when it is accessible and diagnostic

(Bone 1995; Feldman and Lynch 1988; Herr et al. 1991). Referrals are accessible in that they are

easy to retrieve, mostly because of their vividness (Herr et al. 1991). Referrals also have a high

diagnostic value because, unlike advertising, they are transmitted by a non-commercial source,

and therefore are generally seen as credible and trustworthy (Bone 1995).

Third, research on consumer herd behavior and informational cascades suggests that

customers exposed to WOM referrals tend to disregard their own information and follow the

choices made by others (Bikhchandani et al. 1998). This behavior leads to suboptimal outcomes

(Banerjee, 1992). Customers believe that other customers have informational advantages about

the product (Huang and Chen 2006), and may even use simple heuristics to drive their decision

such as “if everyone tells me to watch this movie, I should go see it, because it should be a great

movie.” Therefore, customers infer product quality prior to purchase, and hence form

expectations, from others’ choices and evaluations.

As argued above, we expect that through these mechanisms, referred customers are more

likely to have higher expectations, end up being disappointed, and are less satisfied with their

movie experience than non-referred customers. Formally, we predict:

H1: Customers exposed to WOM referrals are, on average, less satisfied with their product

or service consumption experience than customers not exposed to WOM referrals.


Moderating role of the match of the referrals with the recipient’s preferences

While we expect a negative main effect of WOM referrals on satisfaction, it is possible that

some of the customers who are exposed to WOM referrals turn out to be more satisfied with their

product choice than others. One important factor that can moderate the relationship between the

exposure to WOM referrals and satisfaction is the degree to which a referral’s recipient receives

referrals that are well-matched to his or her preferences.

In contrast to companies and other untargeted sources of information, the referring customer

tends to have a good knowledge about the referred customer’s tastes, making his or her referrals

well-matched with the recipient’s preferences (Dichter 1966). As Duhan et al. (1997) argue,

influential recommendations require knowledge of both the product and the person receiving the

recommendation, especially for products with affective evaluative cues (subjective criteria such

as taste), such as movies. Strong ties, for example close friends, are likely to be knowledgeable

of each other’s tastes and the relevance of their information (Brown and Reingen 1987; Duhan et

al. 1997). Moreover, homophily, the tendency for people to interact with people like them,

suggests that customers are likely to share their experience with people that are similar to

themselves (Brown and Reingen 1987; McPherson et al. 2001; Schmitt et al. 2011; Van den

Bulte and Wuyts 2007). The homophily in taste between these customers is likely to increase the

odds of satisfaction on the recipient side.

We expect that the extent to which customers receive well-matched referrals will moderate

the negative effect of exposure to WOM referrals on satisfaction. In particular, referred

customers who typically receive well-matched referrals are expected to show a higher level of

satisfaction than customers who typically receive ill-matched referrals. Formally:

H2: The degree to which a referral’s recipient receives well-matched referrals moderates

the relationship between the exposure to WOM referrals and satisfaction, in that customers who

are exposed to referrals and receive well-matched referrals are more satisfied with the product

or service consumption experience than customers who receive ill-matched referrals.

Impact of satisfaction on WOM referral value

A second condition for satisfaction to be mediating the effect of WOM referral exposure and

the referral match moderator on WOM referral value is that there also exists an effect of

satisfaction on WOM referral value. Customer satisfaction with a consumption experience is


regarded as a key antecedent of product- or service-related WOM (Anderson 1998; De Matos

and Rossi 2008; Westbrook 1987). The higher the satisfaction of an individual with a

consumption experience, the greater the amount of recommendations she is likely to make.

Several reasons for this effect are captured in the utility-based model proposed by Anderson

(1998), who asserts that a customer’s utility of referring a product of service increases as the

satisfaction with product experience increases.

First, extreme information is more accessible than moderate information (Anderson 1998).

A very satisfying product experience is more memorable and thus more likely to be talked about

than a less satisfying product experience. Moreover, extreme information has been found to be

more diagnostic: an extreme cue provides more utility in discriminating between alternative

categories than a less extreme cue (Anderson 1998; Skowronski and Carlston 1989; Skowronski

and Carlston 1987). As such, higher satisfaction has a strong impact on WOM behavior as it

allows customers to easily categorize the product in the “talk about” compared to “not talk

about” category while moderate satisfaction makes categorization ambiguous.

Second, customers often engage in WOM referrals to bring back positive feelings to relive

the pleasurable product experience and elicit positive feelings (Berger 2014). Dichter (1966)

refers to WOM referrals as having the function of “verbal consumption.” A more satisfying

consumption experience is therefore more likely to be talked about for these purposes as it is

surrounded by more positive feelings that can be retrieved.

Third, a common motive for customers to engage in WOM referrals is self-enhancement

(Berger 2014; De Angelis et al. 2012), where customers talk about favorable and interesting

things to look good in front of their peers and create good impressions. Customers can get

positive recognition from others by linking these favorable impressions with themselves by

talking about them. For this purpose, the more positive a consumption experience is perceived by

the customer, the more it makes the customer look interesting in front of their peers, making it

more suitable to talk about for self-enhancement purposes.

All in all, these motives suggest that the more satisfied a customer is, the more likely she

will share her experience with others. A recent meta-analysis by De Matos and Rossi (2008)

gives strong empirical support for this positive relationship. While we do not formally

hypothesize the relationship between satisfaction and referral value, we aim to empirically verify

the findings above with a moderated mediation analysis.


Data and Variables

In order to test our hypotheses, we ran a large-scale survey in August 2014 among movie

viewers in the United States. The survey investigated the WOM referral behavior of the

respondents and contained a number of psychometric scales necessary to test our hypotheses. In

order to control for inherent differences that exist between the movies watched by the

respondents, we combined the survey with data on movie characteristics that we extract from the

Internet Movie Database (IMDb; www.IMDb.com).

Our choice to use a survey is consistent with current practice in the WOM literature

(Anderson 1998; File et al. 1994; Neelamegham and Jain 1999; Uncles et al. 2013; Von

Wangenheim and Bayón 2006; Westbrook 1987; Yang et al. 2012). In light of our research

objectives, our survey items have multiple benefits over other data types such as social network

data (Goel et al. 2012; Leskovec et al. 2007), or aggregate time-series (Trusov et al. 2009;

Villanueva et al. 2008). First, they allow us to measure individual psychometric constructs (e.g.,

satisfaction with a consumption experience) that are latent and difficult to measure with

secondary data. Second, they provide us the instruments we need to handle the issue of

endogenous selection. As the selection mechanism operates on the individual level—customers

who are exposed to WOM referrals are systematically different from those who were not—it is

challenging for aggregate time-series data to capture individual-level selection effects. Large-

scale social network data, albeit on the individual level, suffers from similar concerns as finding

instrumental variables is extremely challenging, although (costly) field experiments or (difficult

to find) natural experiments may be employed (Aral and Walker 2014). As survey data is prone

to measurement error and recollection bias, which may limit the generalizability of the study, we

use a structural equation model with a measurement component in order to account for

measurement error (Bagozzi and Yi 2012; Kline 2011). We also check that our results do not

suffer from a potential recollection bias in the “Robustness Checks” section.


Measures and variable operationalization

Participants were recruited using the Amazon Mechanical Turk (MTurk) service.3 Workers

were allowed to participate if they had seen a movie in a movie theater in the past 12 months.4 In

order to receive $.5 payment after completing the survey, respondents received a randomized

code to enter in the MTurk environment. The survey consisted of two parts. First, participants

were asked to recall their last movie visit (for a recent paper using a similar survey see Yang et

al. 2012). In this part of the survey, participants reported the title of the last movie they saw at a

movie theater, and were asked to answer a set of questions with this movie in mind (see Table 2

for an overview of the items and scales):

- WOM referral exposure. Respondents answered for the movie they indicated as the last

movie they had seen, whether they were exposed to WOM referrals prior to seeing the

movie. This binary variable takes value one for respondents who reported being exposed to

WOM referrals and zero for those who were not.

- Satisfaction. The degree of satisfaction of the respondent with his/her movie of choice was

measured with three items (one reversed) and seven-point Likert scales, adopted from

Maxham III and Netemeyer (2002).

- WOM referral value. We capture WOM referral value by measuring the total number of

referrals a customer has made and plans to make in the future. To account for the long-tailed

distribution, we take the natural logarithm of the total of number of prospective customers a

consumer referred to and intends to refer in the future, plus one to account for zero values.

In the second part of the survey, participants were asked to complete items that were not specific

to the movie in question:

- Referral match. We measure the degree to which a customer receives referrals that are well-

matched to his or her preferences using three items and seven-point scales. Participants were

asked how they would qualify the movie recommendations they receive in general, how

satisfied they are with the movie recommendations they receive in general, and how well the

individuals who generally recommend them movies know the participant’s movie tastes.

3 For a recent empirical application of MTurk in marketing see for example Lamberton and Rose (2012). For discussions on the MTurk platform and expected reliability of the data see Buhrmester et al. (2011) and Peer et al. (2014). 4 Following recommendations by Peer et al. (2014), U.S. workers with at least a 95% acceptance rate and at least 500 accepted tasks were sampled. Moreover, we applied the procedure outlined by Peer et al. (2012) to prevent sampling multiple responses per worker ID.


These items capture the extent to which, in general, the referrals a participant receives

match her preferences. Because we consider this construct as a moderator, rather than a

mediator, we measure referral matching using a construct not specific to the focal movie

consumption episode, but as an individual-level trait that is determined independently of the

current consumption episode.

We also collected additional individual-level variables that serve as control variables and

instrumental variables in the model (see “The Model” section for how all variables enter the

model):

- Opinion seeking. The scale was adopted from Flynn et al. (1996) and captures the extent to

which a customer looks for opinions from others before choosing movies with six seven-

point Likert items (three reversed).

- Opinion leadership. The scale was adopted from Flynn et al. (1996) and captures the extent

to which the customer exerts influence on the movie choices of others with six seven-point

Likert items (three reversed).

- Gender and age. Participants were also asked about their gender (male coded as 1, female as

0) and age (continuous scale).

Finally, we also collected movie-level data on the movies from IMDb:

- Movie rating. We control for quality differences between movies using the movie ratings

provided by IMDb users.

- Opening weekend box-office revenues. This variable measures the movie revenues up to the

opening weekend and hence captures the popularity of the movie early in its lifecycle. We

use the natural logarithm of this variable to account for its long-tailed distribution.

Sample and descriptive statistics

We collected data from 900 respondents. We excluded respondents who claimed to have

seen the movie before the release date shown on IMDb and respondents who reported a movie

which actual release date, according to IMDb, was before 2012 (because of the short lifecycle of

movies, it is unlikely that these customers saw the movie in the theater). We also removed

duplicate IP addresses, respondents for which we could not match the self-reported movie title

with the IMDb repository, as well as movies for which there was missing data.


Our final sample consists of 851 respondents. Among the respondents, 320 (about 38%)

mentioned having been exposed to WOM referrals prior to seeing the movie, the remaining 531

(about 62%) were not exposed to WOM referrals. Table 3 shows descriptive statistics. In line

with previous research, customers who indicated to be exposed to WOM referrals seem to make

more referrals, compared to customers who were not exposed to WOM referrals (T(849) = -6.824,

p < .001) when we do not control for any other variables. Moreover, when we do not control for

endogeneity, customers who were exposed to WOM referrals seem to show a higher satisfaction

and referral value than those who were not exposed to WOM referrals (e.g., satisfaction 1: T(849)

= -5.062, p < .001). Another interesting difference is that customers who were exposed to WOM

referrals have a higher tendency to be opinion seekers (e.g., opinion seeking 1: T(849) = -6.905, p

< .001). Customers exposed to WOM are also slightly younger (T(849) = 2.680, p = .008), but both

groups do not differ in the proportion of men/women (T(849) = -.057, p = .954). In terms of their

movie choice, customers exposed to WOM tend to choose better rated movies (T(849) = -9.414, p

< .001) and movies with higher opening weekend box-office revenues (T(849) = -4.127, p < .001).

Measurement properties

We first performed a confirmatory factor analysis measurement model to validate and purify

the multi-item scales (Anderson and Gerbing 1988). The items of the latent constructs

satisfaction, referral match, opinion seeking and leadership scales load on separate factors with

free covariances between factors. No cross-loadings were allowed. For each latent variable, we

specify for respondent i, and indicator j:

yij = νj + λjηi + εij, (1)

where ηi is the latent variable, yij is the observed indicator, νj is an intercept, λj is a loading, and

error term εij ~ N(0, σεij). We fixed one loading of each factor to one for identification purposes,

and allow for within-factor covariances between errors of negatively (reversed) scaled items to

parsimoniously control for response styles of negatively worded items (DiStefano and Motl

2006; Marsh 1996; Tomás and Oliver 1999). After dropping the first opinion leadership item

with a poor standardized loading (.41; see Table 2), the measurement model showed adequate fit

(χ2 = 406.671; Comparative Fit Index (CFI) = .962; Tucker-Lewis Index (TLI) = .953; Root

Mean Squared Error of Approximation (RMSEA) = .057; Standardized Root Mean Square

Residual (SRMR) = .045).


Table 4 shows the Average Variance Extracted (AVE), Composite Reliability (CR) and the

correlations between the latent variables (Bagozzi and Yi 2012; Fornell and Larcker 1981). The

results show that the factors are reliable, as shown by high CRs. We assess discriminant validity

by means of the three Fornell and Larcker (1981) criteria. First, the correlations between latent

variables are clearly less than one. Second, the AVEs of all latent variables exceed 50%. Third,

the AVEs of all latent variables exceed the variance shared with other latent variables (squared

correlations). Using the same criterion, we also check discriminant validity between the factors

and the single indicator variables and find that satisfaction and WOM referral value (ρ2 = .284),

and opinion leadership and WOM referral value (ρ2 = .125) are discriminant valid given that the

respective AVEs are larger than the squared correlations. Overall, we conclude that we have

reliable and discriminant valid measures.

The Model

We use a structural equation model (SEM) to investigate the drivers of WOM referrals,

augmented with the measurement model in Equation 1. SEM appropriately deals with survey

data accounting for measurement error (Bagozzi and Yi 2012; Kline 2011) and allows us to

perform the moderated mediation analysis (Figure 2) while accounting for the endogenous

selection mechanism. Below, we describe our model and explain how we account and test for

endogeneity.

Model specification

We specify the following structural equations:

WOMreferralexposurei = αo + α1Opinionseekingi + (2)

α2 ln(Openingweekendboxofficerevenuei) + α3Movieratingi + α4Genderi + α5Agei + ε1,i

Satisfactioni = β1WOMreferralexposurei + β2Referralmatchi + (3)

β3WOMreferralexposurei ∗ Referralmatchi + β4Movieratingi + β5Genderi + β6Agei + ε2,i

WOMreferralvaluei = γ0 + γ1WOMreferralexposurei + γ2Satisfactioni + (4)

γ3Opinionleadershipi + γ4Movieratingi + γ5Genderi + γ6Agei + ε3,i


with the errors ε1,i, ε2,i, ε3,i ~ N(0, 0, 0; Σ). The specification of the error covariance matrix is

defined in the next subsection. Consistent with the mediation literature, we specify a direct effect

of WOM referral exposure on WOM referral value (Preacher et al. 2007; Zhao et al. 2010).

Equation 2 is specified as a linear probability model (Olsen 1982). The intercept in the

satisfaction equation is suppressed for identification purposes. We estimate the measurement

model and structural model simultaneously in Mplus (Muthén and Muthén 1998-2012) using

full-information robust (to deviations from normality) maximum likelihood.5 We follow the

procedure by Klein and Moosbrugger (2000) to estimate the latent interaction effect in Equation

3.

Accounting for endogenous selection

A key feature of our approach is that we consider the endogenous process by which

customers select themselves into being exposed to WOM referrals. Specifically, we argue that

the effects of WOM referral exposure are likely to be driven by endogenous selection on

unobservables. Two types of endogeneity are likely to occur.

First, some customers may be more likely to activate others to discuss their consumption

experiences, while simultaneously being more inclined to seek social contacts. For example,

some customers may simply be more social in that they often talk with other customers about

their favorable movie experiences, but are also exposed to movie-related conversations from

their peers. Moreover, some customers may simply be more involved with the product category.

They like to talk about movies with others and do not hesitate to ask for interpersonal advice

when they need information about a movie. Second, some products are simply more likely to be

talked about (Berger and Schwartz 2011) leading to a potentially spurious correlation between

WOM input and output. For example, some movies are more provocative than others, such that

they generate a lot of chatter.

To correct for endogeneity, we use an instrumental variable procedure and specify a WOM

referral exposure equation (Equation 2) as a function of a number of instruments, and free off-

diagonal elements of the Σ covariance matrix of the structural error terms (Skrondal and Rabe-

Hesketh 2004). We use two instruments, opinion seeking to account for the potential selection at

5 Given that we use robust maximum likelihood to estimate the model, we rescale the test statistic using the scaling correction factor described by Satorra (2000) when performing nested model (likelihood ratio) comparison tests.


the customer level, and opening weekend box-office revenue (in natural logarithm) to account

for the potential selection at the movie level. We demonstrate the statistical properties of these

instruments in the next section. Our choice for these instruments is justified by the fact that we

expect opinion seekers to show a high tendency to seek information from their peers when

buying a new product or service, and hence to be exposed to WOM referrals (Flynn et al. 1996).

Moreover, we expect that the better a movie performs during the first weekend, the more buzz it

will generate and the larger the base of adopters who will recommend the movie to others after

the first weekend.

Testing the instruments

We use a three-pronged approach to empirically test for the strength and validity of our

instruments. For the strength, Stock et al. (2002) provide a rule of thumb in that the F-statistic on

the excluded instruments in the first stage should be greater than ten. A joint test of the R2

increase (from .099 to .143) of the excluded instruments yields F(2, 845) = 21.69, while both

instruments significantly explain WOM referral exposure (p = .043 and p = .001), indicating that

the instruments are empirically strong.

For the validity, we test for overidentifying restrictions (Kenny and Milan 2012). For each

equation, we sequentially free paths from one of the instruments to the dependent variables

(satisfaction and WOM referral value) in the model. The additional paths were not significant

(highest Z = -.696, p = .487), and adding them did not significantly increase model fit (highest χ2

= .473, p = .492), indicating that the instruments are empirically valid.

Finally, we also tested for endogeneity using the Durbin-Wu-Hausman (DWH) test. We find

that the residual is significant in the satisfaction equation (Z = 8.671, p = .003) whereas it is not

significant in the WOM referral value equation (Z = 1.408, p = .235). We conclude that the

instruments are strong and valid and that WOM referral exposure is endogenous in the

satisfaction equation (see Dinner et al. 2014, for a similar procedure).

Moderated mediation analysis

As shown in Figure 2, the expected impact of WOM referral exposure on referral value is

not only mediated by satisfaction, but its effect on the satisfaction mediator is also expected to be

moderated by referral match. As such, we expect a moderated mediation (Preacher et al. 2007).

Consistent with the mediation literature and our model, we decompose the total effect of


exposure to WOM referrals on post-consumption WOM value in a conditional indirect

(mediation) effect and a direct effect (Preacher et al. 2007; Zhao et al. 2010). We specify the

indirect effect, which is conditional on a value of the referral match moderator, as:

Indirect effect = γ2(β1 + β3 ∗ Referralmatch), (5)

while the direct effect is simply γ1. The total effect is the sum of both effects:

Total effect = γ1 + γ2(β1 + β3 ∗ Referralmatch). (6)

The total effect informs us whether customers exposed to WOM referrals differ in their referral

value from the customers not exposed to WOM referrals. The indirect effect allows to test

whether satisfaction mediates the relationship between WOM referral exposure (and the referral

match moderator) and referral value. Because the indirect effect is conditional on the value of the

moderator, we report these effect for three values of the moderating referral match: a low value

equal to the mean minus two standard deviations, the mean value, and a high value equal to the

mean plus two standard deviations. We report the 95% bootstrapped (10,000 iterations)

confidence intervals (CI) around these effects (Preacher et al. 2007; Zhao et al. 2010). We also

compute the proportion mediated (Alwin and Hauser 1975; MacKinnon et al. 2007), which is

given by:

Proportion mediated =|Indirect effect|

|Indirect effect| + | Direct effect| . (7)

Results

We estimate different variants of our model: (i) Model 1 that does not account for

endogenous selection, nor mediation (Equation 4 only), (ii) Model 2 that specifies all three

equations, but does not control for endogeneity (i.e., the errors are not correlated), (iii) Model 3,

our full model accounting for endogeneity (i.e., the errors are now correlated) and mediation.

Parameter estimates of these models are reported in Table 5. Below, we first investigate whether

exposure to WOM referrals influence the referral value of customers. We then show that not

accounting for endogeneity would have substantially changed the answer to this question.

Finally, we explore the mechanism linking exposure to WOM referrals and a customer referral

value by testing each of the hypotheses.


Does exposure to WOM referrals influence customers’ referral value?

Table 6 depicts the estimated total effect of WOM referral exposure on a customer referral

value according to Model 3 and decomposes it into a direct and an indirect effect. A graphical

representation of this total effect decomposition is provided in Figure 3. The total effects (darkest

bars) show that customers who received a movie referral do not have a higher probability to refer

the movie to others than customers who did not receive a referral. In fact, for mean and high

values of the referral match moderator, the total effect of WOM referrals on WOM referral

value, taking into account all paths, is negative yet indistinguishable from zero (95% bootstrap

CI (-.701; .073) for mean referral match and (-.586; .263) for high referral match both contain

zero). However, for a low value of the moderating variable referral match, the total effect

becomes significantly negative (95% bootstrap CI (-.877; -.051) does not contain zero). These

results provide us a first important managerial insight: customers exposed to WOM referrals do

not have a higher referral value than the customers not exposed to WOM. At best, they refer the

movie they have seen to as many other customers as the customers not exposed to WOM

referrals. In some cases, they tend to refer it to less people than the customers who were not

exposed to WOM referrals.

This total effect can be decomposed into a direct effect (mid-grey bars) and an indirect

(mediating) effect via satisfaction (light grey bars). As shown in Figure 3, the effects are of

opposite signs, what Zhao et al. (2010) name “competitive mediation”, and also referred to as

“inconsistent mediation” (MacKinnon et al. 2007). We find a negative indirect effect of WOM

referral exposure on WOM referral value, mediated by satisfaction. The indirect effect is

significant for low and mean values of referral match (95% bootstrap CIs resp. (-1.013; -.224)

and (-.850; -.105) do not contain zero). For high values of referral match, we find an indirect

effect indistinguishable from zero (95% bootstrap CI (-.727; .093) contains zero). Contrasting

with this negative indirect effect, we find a significant direct effect (95% bootstrap CI (.022;

.294) does not contain zero) of WOM referral exposure on WOM referral value. While

satisfaction partially mediates the relationship between WOM referral exposure and a customer

referral value, the proportion mediated for low, mean, and high values of referral match is

respectively .789, .736, and .648, suggesting that a majority of the total effect of WOM referral

exposure on WOM referral value comes from the indirect effect. Following Kenny (2014), this is

close to a full mediation as the share approaches 80%. These results provide us a second


important managerial insight: satisfaction plays a major role in the extent to which customers

who receive a movie referral would pass it on to others and therefore be of high referral value.

Finally, Figure 3 offers an interesting contrast between customers who typically receives ill-

matched referrals (left panel of Figure 3, “low referral match”), and customers who usually

receives well-matched referrals (right panel of Figure 3, “high referral match”). The indirect

effect of exposure to WOM referrals via satisfaction is less pronounced for the second segment

of customers compared to the first segment of customers. In particular, the first segment ends up

recommending the movie less to others when they go for a movie they have been recommended

to than when the movie was not recommended to them. For the second segment, there is no

difference in the number of referrals they make whether the movie has been referred to them or

not. The results provide us a third important managerial insight: the satisfaction of referred

customers who receive well-matched movie referrals is not sufficiently high to make their

referral values higher than the referral value of customers who do not receive any referral.

How do the results change when not accounting for endogeneity?

To highlight the role of endogeneity, we contrast these results with the results from Model 1,

which does not control for it. In Table 5, we find a positive and significant effect of WOM

referral exposure on WOM referral value (γ1 = .256, p = .001, 95% bootstrap CI (.109; .404)

does not contain zero). This result is consistent with previous research and indicates that the

apparent positive correlation between receiving and giving WOM referrals is due to endogenous

self-selection rather than the exposure to WOM itself.6

What explains the effect of exposure to WOM referrals on customers’ referral value?

Our model also provides insights into the mechanisms that relate exposure to WOM referrals

and WOM referral value, which allow us to test the hypotheses (see Table 5, Model 3). Based on

the moderated mediation analyses we find, consistent with H1 and H2, that the effect of WOM

referral exposure on referral value is moderated by the referral match and mediated by

satisfaction. We find a negative and significant main effect of WOM referral exposure on

satisfaction (β1 = -.957, p = .013). We explain this negative effect by the fact that customers who

6 Note that the change in the size of the standard errors (efficiency loss) between Model 1 and Model 3 (approximately a factor two) is in line with previous research (Kornish and Ulrich 2014; Zhang et al. 2009), which we consider reasonable.


received a movie referral tend to form higher expectations about the movie they are about to see,

increasing the odds that they will end up disappointed and hence less satisfied compared to

customers who did not received a referral. This effect is positively moderated by the referral

match (β3 = .233, p = .057), consistent with H2. Customers who tend to receive referrals that fit

their preferences end up more satisfied with the movie than those who receive ill-matched

referrals. When the referring customer has a good knowledge about the movie preferences of the

referred customer, for example in case of strong ties, we find that the negative effect of receiving

a referral becomes smaller in intensity. As shown in Figure 3, the indirect effect becomes

insignificant for high values of referral match.

To complete the mediating effect of satisfaction, the results also show a positive and

significant effect of satisfaction on WOM referral value (γ2 = .455, p < .001). Customers who are

more satisfied get more utility from engaging in WOM referrals themselves and spread the word

to a larger audience (Anderson 1998; De Matos and Rossi 2008). Finally, we find a small but

significant positive direct effect of WOM referral exposure on customer referral value (γ1 = .156,

p = .024) as we described in the prior subsection. Although the direct effect accounts for a small

share of the total effect, it may capture several interesting mechanisms, which we do not identify

in our study. For example, the theory of emotional contagion shows that emotions can flow from

one person to another (Hennig-Thurau et al. 2006). Arousal that drives customers to engage in

product-related conversations (Berger 2011) can be transferred to the referral recipient, leading

her to refer in turn. Alternatively, customers may engage in indirect reciprocity after receiving a

favorable referral in that they reciprocate the act of receiving a successful referral by referring to

others in their network (Wasko and Faraj 2005). Future research could shed more light on the

nature of this small but significant positive effect by introducing additional mediators or

moderators (Zhao et al. 2010).

Robustness checks

We check the robustness of these results to various scenarios. First, to investigate potential

recollection bias, we restrict the sample to customers who saw the movie less than three months

ago, resulting in a sample size of 660 customers (77.6% of the full sample). We find the

coefficients of interest and the substantive conclusions to be identical.


Second, the literature suggests that early adopters might differ from late adopters (Rogers

1983), and such differences can have an effect of their referral value. We control for the timing

of adoption (viewing) of the movie and include a variable that captures how many months after

the movie release a respondent saw the movie to all three equations. We find that the additional

parameters are insignificant in all equations (all p > .344), that the coefficients of interest remain

substantively identical, and that the Bayesian Information Criterion (BIC) becomes larger (BIC =

45,785 with adoption timing compared to 45,766 without adoption timing).

Third, we checked the robustness of our results when allowing for a non-zero error

covariance between Equations 2 and 4, in addition to the error covariance between Equations 2

and 3. We find that the quantities of interest are substantively identical and that the additional

error covariance is virtually zero (σ1,3 = -.008, p = .906), and that the BIC becomes larger (BIC =

45,773).

Fourth, we account for the possibility that the direct effect of WOM referral exposure on

WOM referral value may also be moderated by referral match. We add a main effect of referral

match and its interaction with WOM referral exposure to Equation 4 and find that both paths are

insignificant (resp. p = .319, p = .202), and that the model fit does not improve (BIC = 45,774).

Discussion

Our paper reveals a number of interesting findings about the relationship between receiving

and giving WOM referrals. First, our results highlight the consequences of ignoring the

endogenous selection process by which customers are exposed to WOM referrals. In contrast to

previous research (Sheth 1971; Trusov et al. 2009; Uncles et al. 2013; Villanueva et al. 2008;

Von Wangenheim and Bayón 2006), the results show that, on average, being exposed to WOM

referrals does not affect the number of referrals a customer will make in turn.

Second, we get insights in the mechanisms that lead to these effects. The moderated

mediation analysis reveals that referred customers who receive referrals that are poorly matched

turn out to be less satisfied than non-referred customers, leading them to refer less in turn. This

result supports the idea that not all referrals are equally useful. Effective recommendations

should take the recipient into account and require knowledge of both the product and the person

receiving the recommendation (Duhan et al. 1997). Consistent with a disconfirmation of

expectations account (Oliver 1980), WOM referrals create unrealistic expectations about the


movie a customer is about to view which are disconfirmed when seeing the movie, leading to

lower satisfaction. The mediation of satisfaction accounts for the majority (for average

customers: 73.6%) of the total effect of WOM referral exposure (moderated by referral match)

on WOM referral value. These results contrast with previous research suggesting that customers

acquired through WOM might be satisfied customers (Bolton et al. 2004; Uncles et al. 2013;

Von Wangenheim and Bayón 2006).

Managerial implications

Our results offer a number of important managerial implications for firms interested in

WOM acquisition strategies. First, while WOM acquisition strategies have certainly shown their

potential in attracting higher CLV customers compared to other acquisition strategies (Schmitt et

al. 2011; Villanueva et al. 2008), they are less likely to recruit customers with a higher referral

value (Kumar et al. 2010). On average, we find that customers acquired via referrals do not have

significantly different referral value than other customers. From a managerial viewpoint, it

implies that firms can count on high revenues from customers recruited via referrals but should

expect more limited cascades of referrals (i.e., the number of customers recruited) than what

prior work has suggested (e.g., Trusov et al. 2009; Villanueva et al. 2008)

Second, our moderated mediation analysis pinpoints that companies should act with care

when implementing WOM acquisition campaigns. It is critical not to encourage referrals

between customers who know little about each other’s preferences. Incentivizing customers to

increase the number of referrals they make, for example by offering them a monetary reward

(Verlegh et al. 2013), might have a detrimental effect on the outcome as it might push customers

to pay less attention to who they share their recommendation with. A solution can be to give

customers incentives to share that are conditional on the adoption of the receiver and/or her

future referral value (Garnefeld et al. 2013; Schmitt et al. 2011).

Third, the results suggest directions for firms on how to boost the length of referral cascades.

One can be to offer technology and software solutions to their current customers that help them

improving the satisfaction of the referred customers, and consequently their referral value.

Different types of helping tools can be made available to them. For instance, information tools

that provide a sensible description of the product and ensure realistic expectations about the

referred products can be made available on a (online) customer platform. For physical products,


free samples can be offered to share, while for services, informational videos can be provided to

referring customers to show to their friends. In addition, firms could also offer matching tools

that help the referring customers to better identify which of their friends would be potentially

most satisfied with the referred product or service. For example, movie enthusiast websites (like

IMDb), but also social networks such as Facebook, often allow users to specify their favorite

movies or show the most recent ratings and reviews for each user when accessing their profile,

effectively showing the movies they have seen and their ratings. Companies may even aid

recommenders by using and analyzing historical data on customer preferences combined with

demographic information, and come up with a list of friends that might also like the movie they

have seen. In view of the increasing amount of data (sources) becoming available, matching tools

undoubtedly have the potential to help customers to better target their recommendations.

Limitations and directions for future research

Our study suffers from several limitations that offer opportunities for future research. First,

our study contributes to a growing stream of research that investigates the consequences of

WOM referrals on future referrals (see Table 1). Similar to these papers, we focus on the

consequences of positive WOM referrals. Studying the consequences of negative WOM was

beyond the scope of the present research but can certainly reveal interesting new insights.

Second, we investigate the effect of WOM referrals on a customer referral value in general.

It would be interesting to study the extent to which the content of these referrals also plays a role.

Future research could distinguish between, for example very enthusiastic vs. mild endorsements,

solicited referrals vs. unsolicited referrals (Fitzsimons and Lehmann 2004), or rewarded vs. not-

rewarded referrals (Verlegh et al. 2013).

Third, like many studies on WOM referrals, we use self-reported data. A key reason for us

to use survey data is that it allows us to study the process: the moderating and mediating

mechanisms that jointly link receiving and giving referrals. While we control for potential

measurement issues with a measurement model, the retrospective nature of the survey might

limit the generalizability of our findings. Future research may augment similar survey data with

secondary data on actual referral activity to further alleviate the potential recollection bias

concerns and enhance the generalizability of the study.


Fourth, we reveal satisfaction to be a key factor in the process that leads to an effect of

WOM referral exposure on WOM referral value. However, we do not capture complete evidence

for the theoretical mechanisms we propose, which are based on expectations. Although the

relationship between expectations and satisfaction is well-established (Oliver, 1980; Westbrook,

1987; Zeithaml et al., 1993), future research may be able to dig even further in the process,

especially by showing an effect of WOM referral exposure on expectations, for example by

means of lab experiments or multiple measures over time.

To conclude, we believe, in view of previous work and the current research, that WOM

acquisition strategies can be potentially successful strategies for firms to consider but should be

implemented with care if they want to benefit from long cascades of referrals.


References

Alwin, Duane F. and Robert M. Hauser (1975), "The Decomposition of Effects in Path Analysis," American Sociological Review, 40 (1), 37-47.

Anderson, Eugene W. (1998), "Customer Satisfaction and Word of Mouth," Journal of Service

Research, 1 (1), 5-17.

——— and Linda C. Salisbury (2003), "The Formation of Market‐Level Expectations and Its Covariates," Journal of Consumer Research, 30 (1), 115-24.

Anderson, James C. and David W. Gerbing (1988), "Structural Equation Modeling in Practice: A Review and Recommended Two-Step Approach," Psychological Bulletin, 103 (3), 411-23.

Aral, Sinan and Dylan Walker (2014), "Tie Strength, Embeddedness, and Social Influence: A Large-Scale Networked Experiment," Management Science, 60 (6), 1352-70.

Bagozzi, Richard P. and Youjae Yi (2012), "Specification, Evaluation, and Interpretation of Structural Equation Models," Journal of the Academy of Marketing Science, 40 (1), 8-34.

Banerjee, Abhijit V. (1992), "A Simple Model of Herd Behavior," The Quarterly Journal of

Economics, 107 (3), 797-817.

Berger, Jonah (2011), "Arousal Increases Social Transmission of Information," Psychological

Science, 22 (7), 891-93.

——— (2014), "Word of Mouth and Interpersonal Communication: A Review and Directions for Future Research," Journal of Consumer Psychology, 24 (4), 586-607.

——— and Eric M. Schwartz (2011), "What Drives Immediate and Ongoing Word of Mouth?," Journal of Marketing Research, 48 (5), 869-80.

Bikhchandani, Sushil, David Hirshleifer, and Ivo Welch (1998), "Learning From the Behavior of Others: Conformity, Fads, and Informational Cascades," Journal of Economic Perspectives, 12 (3), 151-70.

Bolton, Ruth N., Katherine N. Lemon, and Peter C. Verhoef (2004), "The Theoretical Underpinnings of Customer Asset Management: A Framework and Propositions for Future Research," Journal of the Academy of Marketing Science, 32 (3), 271-92.

Bone, Paula F. (1995), "Word-of-Mouth Effects on Short-term and Long-term Product Judgments," Journal of Business Research, 32 (3), 213-23.


Brown, Jacqueline J. and Peter H. Reingen (1987), "Social Ties and Word-of-Mouth Referral Behavior," Journal of Consumer Research, 14 (3), 350-62.

Buhrmester, Michael, Tracy Kwang, and Samuel D. Gosling (2011), "Amazon's Mechanical Turk: A New Source of Inexpensive, Yet High-Quality, Data?," Perspectives on Psychological

Science, 6 (1), 3-5.

Chevalier, Judith A. and Dina Mayzlin (2006), "The Effect of Word of Mouth on Sales: Online Book Reviews," Journal of Marketing Research, 43 (3), 345-54.

De Angelis, Matteo, Andrea Bonezzi, Alessandro M. Peluso, Derek D. Rucker, and Michele Costabile (2012), "On Braggarts and Gossips: A Self-Enhancement Account of Word-of-Mouth Generation and Transmission," Journal of Marketing Research, 49 (4), 551-63.

De Matos, Celso A. and Carlos A.V. Rossi (2008), "Word-of-Mouth Communications in Marketing: A Meta-Analytic Review of the Antecedents and Moderators," Journal of the

Academy of Marketing Science, 36 (4), 578-96.

Dichter, Ernest (1966), "How Word-of-Mouth Advertising Works," Harvard Business Review, 44 (6), 147-60.

Dinner, Isaac M., Harald J. Van Heerde, and Scott A. Neslin (2014), "Driving Online and Offline Sales: The Cross-Channel Effects of Traditional, Online Display, and Paid Search Advertising," Journal of Marketing Research, 51 (5), 527-45.

DiStefano, Christine and Robert W. Motl (2006), "Further Investigating Method Effects Associated With Negatively Worded Items on Self-Report Surveys," Structural Equation

Modeling: A Multidisciplinary Journal, 13 (3), 440-64.

Duhan, Dale F., Scott D. Johnson, James B. Wilcox, and Gilbert D. Harrell (1997), "Influences on Consumer Use of Word-of-Mouth Recommendation Sources," Journal of the Academy of

Marketing Science, 25 (5), 283-95.

Feldman, Jack M. and John G. Lynch (1988), "Self-Generated Validity and Other Effects of Measurement on Belief, Attitude, Intention, and Behavior," Journal of Applied Psychology, 73 (3), 421-35.

File, Karen M., Dianne S.P. Cermak, and Russ A. Prince (1994), "Word-of-Mouth Effects in Professional Services Buyer Behaviour," Service Industries Journal, 14 (3), 301-14.

Fitzsimons, Gavan J. and Donald R. Lehmann (2004), "Reactance to Recommendations: When Unsolicited Advice Yields Contrary Responses," Marketing Science, 23 (1), 82-94.


Flynn, Leisa R., Ronald E. Goldsmith, and Jacqueline K. Eastman (1996), "Opinion Leaders and Opinion Seekers: Two New Measurement Scales," Journal of the Academy of Marketing

Science, 24 (2), 137-47.

Fornell, Claes and David F. Larcker (1981), "Evaluating Structural Equation Models with Unobservable Variables and Measurement Error," Journal of Marketing Research, 18 (1), 39-50.

Garnefeld, Ina, Andreas Eggert, Sabrina V. Helm, and Stephen S. Tax (2013), "Growing Existing Customers' Revenue Streams Through Customer Referral Programs," Journal of

Marketing, 77 (4), 17-32.

Godes, David and Dina Mayzlin (2009), "Firm-Created Word-of-Mouth Communication: Evidence from a Field Test," Marketing Science, 28 (4), 721-39.

Goel, Sharad, Duncan J. Watts, and Daniel G. Goldstein (2012), "The Structure of Online Diffusion Networks," Proceedings of the 13th ACM Conference on Electronic Commerce, 623-38.

Haenlein, Michael and Barak Libai (2013), "Targeting Revenue Leaders for a New Product," Journal of Marketing, 77 (3), 65-80.

Hennig-Thurau, Thorsten, Markus Groth, Michael Paul, and Dwayne D. Gremler (2006), "Are All Smiles Created Equal? How Emotional Contagion and Emotional Labor Affect Service Relationships," Journal of Marketing, 70 (3), 58-73.

Herr, Paul M., Frank R. Kardes, and John Kim (1991), "Effects of Word-of-Mouth and Product-Attribute Information on Persuasion: An Accessibility-Diagnosticity Perspective," Journal of

Consumer Research, 17 (4), 454-62.

Hinz, Oliver, Bernd Skiera, Christian Barrot, and Jan U. Becker (2011), "Seeding Strategies for Viral Marketing: An Empirical Comparison," Journal of Marketing, 75 (6), 55-71.

Huang, Jen‐Hung and Yi‐Fen Chen (2006), "Herding in Online Product Choice," Psychology &

Marketing, 23 (5), 413-28.

Kenny, David A. (2014), "Mediation," (accessed July 15, 2015), [available at http://davidakenny.net/cm/mediate.htm].

——— and Stephanie Milan (2012), "Identification: A Nontechnical Discussion of a Technical Issue," in Handbook of Structural Equation Modeling, Rick H. Hoyle, ed. New York: Guilford Press.


Klein, Andreas and Helfried Moosbrugger (2000), "Maximum Likelihood Estimation of Latent Interaction Effects with the LMS Method," Psychometrika, 65 (4), 457-74.

Kline, Rex B. (2011), Principles and Practice of Structural Equation Modeling. New York: Guilford Press.

Kornish, Laura J. and Karl T. Ulrich (2014), "The Importance of the Raw Idea in Innovation: Testing the Sow's Ear Hypothesis," Journal of Marketing Research, 51 (1), 14-26.

Kumar, V., J.A. Petersen, and Robert P. Leone (2010), "Driving Profitability by Encouraging Customer Referrals: Who, When, and How," Journal of Marketing, 74 (5), 1-17.

Lamberton, Cait P. and Randall L. Rose (2012), "When is ours better than mine? A framework for understanding and altering participation in commercial sharing systems," Journal of

Marketing, 76 (4), 109-25.

Leskovec, Jure, Lada A. Adamic, and Bernado A. Huberman (2007), "The Dynamics of Viral Marketing," ACM Transactions on the Web, 1 (1), 1-39.

MacKinnon, D.P., A.J. Fairchild, and M.S. Fritz (2007), "Mediation Analysis," Annual Review of

Psychology, 58 (1), 593-614.

Marsh, Herbert W. (1996), "Positive and Negative Global Self-Esteem: A Substantively Meaningful Distinction or Artifactors?," Journal of Personality and Social Psychology, 70 (4), 810.

Maxham III, James G. and Richard G. Netemeyer (2002), "A Longitudional Study of Complaining Customers' Evaluations of Multiple Service Failures and Recovery Efforts," Journal of Marketing, 66 (4), 57-71.

McPherson, Miller, Lynn Smith-Lovin, and James M. Cook (2001), "Birds of a Feather: Homophily in Social Networks," Annual Review of Sociology, 27 (1), 415-44.

Murray, Keith B. (1991), "A Test of Services Marketing Theory: Consumer Information Acquisition Activities," Journal of Marketing, 55 (1), 10-25.

Muthén, L.K. and B.O. Muthén (1998-2012), Mplus User's Guide. Los Angeles: Muthén & Muthén.

Nair, Harikesh S., Puneet Manchanda, and Tulikaa Bhatia (2010), "Asymmetric Social Interactions in Physician Prescription Behavior: The Role of Opinion Leaders," Journal of

Marketing Research, 47 (5), 883-95.


Neelamegham, Ramya and Dipak Jain (1999), "Consumer Choice Process for Experience Goods: An Econometric Model and Analysis," Journal of Marketing Research, 36 (3), 373-86.

Oliver, Richard L. (1980), "A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions," Journal of Marketing Research, 17 (4), 460-69.

Olsen, Randall J. (1982), "Distributional Tests for Selectivity Bias and a More Robust Likelihood Estimator," International Economic Review, 23 (1), 223-40.

Peer, Eyal, Gabriele Paolacci, Jesse Chandler, and Pam Mueller (2012), "Selectively Recruiting Participants from Amazon Mechanical Turk Using Qualtrics," working paper (accessed August 15, 2014), [available at http://ssrn.com/abstract=2100631].

———, Joachim Vosgerau, and Alessandro Acquisti (2014), "Reputation as a Sufficient Condition for Data Quality on Amazon Mechanical Turk," Behavior Research Methods, 46 (4), 1023-31.

Preacher, Kristopher J., Derek D. Rucker, and Andrew F. Hayes (2007), "Addressing Moderated Mediation Hypotheses: Theory, Methods, and Prescriptions," Multivariate Behavioral Research, 42 (1), 185-227.

Rogers, Everett M. (1983), Diffusion of Innovations. New York: Free Press.

Satorra, Albert (2000), "Scaled and Adjusted Restricted Tests in Multi-Sample Analysis of Moment Structures," in Innovations in Multivariate Statistical Analysis, Vol. 36, Risto D.H. Heijmans, D.S.G. Pollock and Albert Satorra, eds. Dordrecht, The Netherlands: Springer.

Schmitt, Philipp, Bernd Skiera, and Christophe Van den Bulte (2011), "Referral Programs and Customer Value," Journal of Marketing, 75 (1), 46-59.

Shankar, Venkatesh, Amy K. Smith, and Arvind Rangaswamy (2003), "Customer Satisfaction and Loyalty in Online and Offline Environments," International Journal of Research in

Marketing, 20 (2), 153-75.

Sheth, Jagdish N. (1971), "Word-of-Mouth in Low-Risk Innovations," Journal of Advertising

Research, 11 (3), 15-18.

Skowronski, John J. and Donal E. Carlston (1989), "Negativity and Extremity Biases in Impression Formation: A Review of Explanations," Psychological Bulletin, 105 (1), 131-42.


———, and ——— (1987), "Social Judgment and Social Memory: The Role of Cue Diagnosticity in Negativity, Positivity, and Extremity Biases," Journal of Personality and Social

Psychology, 52 (4), 689-99.

Skrondal, Anders and Sophia Rabe-Hesketh (2004), Generalized Latent Variable Modeling:

Multilevel, Longitudinal, and Structural Equation Models. Boca Raton: CRC Press

Stock, James H., Jonathan H. Wright, and Motohiro Yogo (2002), "A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments," Journal of Business

& Economic Statistics, 20 (4), 518-29.

Tomás, José M. and Amparo Oliver (1999), "Rosenberg's Self‐Esteem Scale: Two Factors or Method Effects," Structural Equation Modeling: A Multidisciplinary Journal, 6 (1), 84-98.

Trusov, Michael, Randolph E. Bucklin, and Koen Pauwels (2009), "Effects of Word-of-Mouth Versus Traditional Marketing: Findings from an Internet Social Networking Site," Journal of

Marketing, 73 (5), 90-102.

Uncles, Mark D., Robert East, and Wendy Lomax (2013), "Good Customers: The Value of Customers by Mode of Acquisition," Australasian Marketing Journal, 21 (2), 119-25.

Van den Bulte, Christophe and Stefan Wuyts (2007), Social Networks and Marketing. Cambridge: Marketing Science Institute.

Verlegh, Peeter W.J. , Gangseog Ryu, Mirjam A. Tuk, and Lawrence Feick (2013), "Receiver Responses to Rewarded Referrals: The Motive Inferences Framework," Journal of the Academy

of Marketing Science, 41 (6), 669-82.

Villanueva, Julian, Shijin Yoo, and Dominique M. Hanssens (2008), "The Impact of Marketing-Induced Versus Word-of-Mouth Customer Acquisition on Customer Equity Growth," Journal of

Marketing Research, 45 (1), 48-59.

Von Wangenheim, Florian and Tomás Bayón (2006), "Satisfaction, Loyalty and Word of Mouth Within the Customer Base of a Utility Provider: Differences Between Stayers, Switchers and Referral Switchers," Journal of Consumer Behaviour, 3 (3), 211-20.

Wasko, Molly M. and Samer Faraj (2005), "Why Should I Share? Examining Social Capital and Knowledge Contribution in Electronic Networks of Practice," Management Information Systems

Quarterly, 29 (1), 35-57.

Westbrook, Robert A. (1987), "Product/Consumption-Based Affective Responses and Postpurchase Processes," Journal of Marketing Research, 24 (3), 258-70.


Yang, Sha, Mantian Hu, Russell S. Winer, Henry Assael, and Xiaohong Chen (2012), "An Empirical Study of Word-of-Mouth Generation and Consumption," Marketing Science, 31 (6), 952-63.

Zeithaml, Valarie A., Leonard L. Berry, and A. Parasuraman (1993), "The Nature and Determinants of Customer Expectations of Service," Journal of the Academy of Marketing

Science, 21 (1), 1-12.

Zhang, Jie, Michel Wedel, and Rik Pieters (2009), "Sales Effects of Attention to Feature Advertisements: A Bayesian Mediation Analysis," Journal of Marketing Research, 46 (5), 669-81.

Zhao, Xinshu, John G. Lynch Jr., and Qimei Chen (2010), "Reconsidering Baron and Kenny: Myths and Truths About Mediation Analysis," Journal of Consumer Research, 37 (2), 197-206.


Tables

Table 1

Comparison of Current Research With Existing Research Studying the Effect of Receiving Referrals on Giving Referrals

Study Industry Type of data Method

Level of

aggregation

Control for

endogenous

selection

Model the

underlying

process Key findings

Sheth (1971) Razor blades Interviews Mean comparisons Individual No No

25% of the respondents who were influenced by a personal source attempted to influence someone else compared to 9% of respondents who were not influenced by a personal source

File, Cermak, and Prince (1994)

Three B2B industries Survey OLS Individual No No Input WOM is positively associated with output WOM for two out of

three industries

Von Wangenheim and Bayon (2006)

Energy providers Survey Mean

comparisons Individual No No Referral switchers (customers who were acquired through positive WOM from other customers) are more likely to make a referral compared to other switchers

Leskovec, Adamic, and Huberman (2007) Online retailing Secondary data Descriptive Individual No No Recommendations rarely successfully propagate

Villanueva, Yoo, and Hanssens (2008) Web hosting Survey and

secondary data VAR Aggregate No No Customers acquired through WOM generate more WOM compared to those acquired through traditional marketing channels

Trusov, Bucklin, and Pauwels (2009)

Online social networking Secondary data VAR Aggregate No No

The long-term elasticity of new sign-ups with respect to WOM is approximately 20 times higher than for marketing events, and 30 times that of media appearances

Goel, Watts, and Goldstein (2012)

Various online services Secondary data Descriptive Individual No No Cascades of subsequent referrals are small

Yang et al. (2012) Car industry Survey Joint utility model Individual No No There is a strong synergy effect between WOM generation and

consumption

Uncles, East, and Lomax (2013)

Various B2C industries Survey Mean

comparisons Individual No No Recommendation rates of customers who were referred are higher than customers acquired through other modes of acquisition

This study Motion-picture industry

Survey and secondary data SEM Individual Yes Yes

On average, customers exposed to WOM referrals do not refer more or less than customers who were not exposed to WOM referrals; customers who receive badly matched referrals have a lower referral value


Table 2

Survey Items and Scales

Construct Item Scale

WOM referral exposure Did anyone recommend you this movie before you saw it?

Yes / No

WOM referral value How many people, approximately, did you recommend the movie to?

#

How many people, approximately, do you intend to recommend the movie to in the future?

#

Satisfaction 1 I am satisfied with my overall experience with the movie

Strongly Disagree - Strongly Agree

Satisfaction 2 As a whole, I am not satisfied with the movie b Strongly Disagree - Strongly Agree

Satisfaction 3 How satisfied are you overall with the quality of the movie?

Very Dissatisfied - Very Satisfied

Referral match 1 In general, how would you qualify the movie recommendations you receive?

Very Bad - Very Good

Referral match 2 In general, how satisfied are you with these movie recommendations?

Very Dissatisfied - Very Satisfied

Referral match 3 In general, how much do they know about your movie preferences? Their knowledge is …

Very Bad - Very Good

Opinion seeking 1 When I consider seeing a movie, I ask other people for advice


Opinion seeking 2 I don't need to talk to others before I see a movie b Strongly Disagree - Strongly Agree

Opinion seeking 3 I rarely ask other people what movies to see b Strongly Disagree - Strongly Agree

Opinion seeking 4 I like to get others' opinions before I see a movie Strongly Disagree - Strongly Agree

Opinion seeking 5 I feel more comfortable seeing a movie when I have gotten other people's opinions on it


Opinion seeking 6 When choosing a movie, other people's opinions are not important to me b


Opinion leadership 1 My opinion about movies seems not to count with other people a b


Opinion leadership 2 When they choose a movie, other people do not turn to me for advice b


Opinion leadership 3 Other people rarely come to me for advice about choosing movies b


Opinion leadership 4 People that I know pick movies based on what I have told them


Opinion leadership 5 I often persuade other people to see movies that I like


Opinion leadership 6 I often influence people's opinions about popular movies


Gender What is your gender? Male / Female

Age What is your age? #

a denotes an item dropped in the final measurement model; b denotes a negatively worded (reversed) item. Note: All "Strongly Disagree - Strongly Agree", "Very Bad - Very Good", and "Very Dissatisfied - Very Satisfied" scales are seven-point.


Table 3

Descriptive Statistics

Complete

sample

(n = 851)

Customers

exposed to

WOM

referrals

(n = 320)

Customers not

exposed to

WOM

referrals

(n = 531)

Variable M SD M SD M SD

ln(WOM referral value) 1.378 1.086 1.696 1.048 1.185 1.064 Satisfaction 1 5.850 1.207 6.116 .984 5.689 1.299 Satisfaction 2 5.378 1.875 5.538 1.913 5.282 1.847 Satisfaction 3 5.945 1.173 6.250 .899 5.761 1.277

Referral match 1 4.155 .893 4.478 .830 3.960 .873 Referral match 2 5.022 .897 5.275 .867 4.870 .882 Referral match 3 5.153 .964 5.416 .926 4.994 .953

Opinion seeking 1 4.041 1.461 4.475 1.353 3.780 1.464 Opinion seeking 2 3.133 1.498 3.484 1.589 2.921 1.400 Opinion seeking 3 3.765 1.539 4.150 1.528 3.533 1.500 Opinion seeking 4 4.118 1.444 4.434 1.417 3.927 1.428 Opinion seeking 5 4.108 1.546 4.522 1.429 3.859 1.562 Opinion seeking 6 3.921 1.535 4.234 1.470 3.733 1.543

Opinion leadership 1 4.859 1.336 4.834 1.388 4.874 1.305 Opinion leadership 2 4.396 1.470 4.509 1.509 4.328 1.444 Opinion leadership 3 4.282 1.527 4.466 1.581 4.171 1.484 Opinion leadership 4 4.363 1.255 4.581 1.219 4.232 1.259 Opinion leadership 5 4.504 1.403 4.828 1.308 4.309 1.423 Opinion leadership 6 4.354 1.364 4.572 1.332 4.222 1.367

ln(Opening weekend box-office revenue) 17.508 1.464 17.772 1.346 17.348 1.509 Movie rating 75.566 9.994 79.522 8.730 73.183 9.958

Gender .402 .491 .403 .491 .401 .491 Age 31.973 9.470 30.856 9.517 32.646 9.387


Table 4

Composite Reliability, Average Variance Extracted, and Correlations Between Factors

CR AVE

Correlations

Opinion

leadership Satisfaction Referral match

Opinion seeking .872 .533 .419 .041 .422 Opinion leadership .858 .550 .152 .342

Satisfaction .771 .530 .292 Referral match .808 .586


Table 5

Parameter Estimates

Model 1 Model 2 Model 3

No endogeneity

correction, no

mediation

No endogeneity

correction, with

mediation

With endogeneity

correction and

mediation

Parameter

estimate SE

Parameter

estimate SE

Parameter

estimate SE

Equation 2: WOM referral exposure Opinion seeking .096*** .014 .095*** .014

ln(Opening weekend box-office revenue) .021** .011 .023** .010 Movie rating .013*** .002 .013*** .002

Gender .038 .032 .038 .032 Age -.002 .002 -.002 .002

Intercept -.957*** .209 -.989*** .197 Equation 3: Satisfaction

WOM referral exposure .112 .076 -.957** .385 Referral match .238*** .086 .315*** .093

WOM referral exposure * Referral match .231* .122 .233* .122 Movie rating .034** .004 .049*** .007

Gender .176** .073 .211*** .081 Age .011*** .004 .007* .004

Equation 4: ln(WOM referral value) Satisfaction .455*** .030 .455*** .030

WOM referral exposure .256*** .075 .157** .069 .156** .069 Opinion leadership .442*** .044 .364*** .038 .365*** .038

Movie rating .025*** .004 .009*** .003 .009*** .003 Gender -.044 .066 -.140** .061 -.140** .061

Age .014*** .003 .008** .003 .008** .003 Intercept -1.071*** .299 -.953*** .302 -1.240*** .323

Error covariance between Equations 2 and 3 .220*** .078 n 851 851 851

Log likelihood -7,420.53 -22,618.19 -22,613.46 Scaling correction factor 1.498 1.363 1.358

Akaike Information Criterion (AIC) 14,887.07 45,394.37 45,386.92 Bayesian Information Criterion (BIC) 14,996.24 45,769.34 45,766.64

# of free parameters 23 79 80 * p < .1; ** p < .05; *** p < .01


Table 6

Decomposition of the Total Effect of WOM Exposure on Customer Referral Value, with 95% Bootstrap CIs

Total effect Indirect effect Direct effect

2.50% M 97.50% 2.50% M 97.50% 2.50% M 97.50%

Low referral match

(M - 2*SD) -.877 -.428 -.051 -1.013 -.584 -.224 .022 .156 .294

Mean referral match -.701 -.279 .073 -.850 -.435 -.105 .022 .156 .294

High referral match

(M + 2*SD) -.586 -.130 .263 -.727 -.287 .093 .022 .156 .294

Note: This table shows the total, indirect (mediation) and direct effects of WOM referral exposure on referral value (based on Model 3 parameter estimates) and their 95% bootstrap confidence intervals (10,000 iterations).


Figures

Figure 1: Visualization of Referral Cascades.

Note: Circles represent customers and arrows indicate referrals. For instance, customer 4

receives a referral from customer 2 and in turn refers to customer 6, 7 and 8.

Time

WOM referral value

for customer 4

WOM referral exposure

for customer 4

Cust.1

Cust.3 Cust.5

Cust.6

Cust.7

Cust.8

Cust.2 Cust.4


Figure 2: Theoretical Framework

H1

H2

Exposure to WOM referrals Satisfaction WOM referral value

Referral match


Figure 3: Decomposition of the Total Effect of WOM Exposure on a Customer Referral

Value

Note: Error bars denote 95% bootstrap confidence intervals.

-1.2

-1.0

-.8

-.6

-.4

-.2

.0

.2

.4Low referral match (M - 2*SD) Mean referral match High referral match (M + 2*SD)

Mar

gina

l effe

ct o

f WO

M re

ferr

al e

xpos

ure

on

ln(W

OM

refe

rral

val

ue)

Total effect Indirect (mediating) effect Direct effect


Acquiring Customers via Word-of-Mouth Referrals: A ...

Documents

Transcript of Acquiring Customers via Word-of-Mouth Referrals: A ...